Abstarct: New technologies make big data that their analysis cause to invent new statistical methods. Before using these methods, some techniques are normally used to reduce the data dimension and provide a basis to quantify the importance of features. The methods such as sure independence screening (SIS) detect the effective variables by ranking. These methods don’t consider the interaction effects, whereas the experience of researcher may indicate the necessity of interactions. In this thesis, by using generalized correlation coefficients, a two-stage algorithm are presented to determine the effective main and interaction effects. The advantage of the proposed method in comparison with SIS is that it can improve the performance of classification techniques, like K-nearest neighbors and centroid baxsed.